Remove exception hooks in favor of real C++ exceptions:
* include/cln/exception.h: New file...
* include/cln/cln.h: ...included here...
* include/cln/rational.h: ...and here.
* include/cln/abort.h: Removed.
* include/cln/integer.h (ash_exception, exquo_exception): Add exception
types.
* include/cln/float.h (floating_point_exception,
floating_point_nan_exception, floating_point_overflow_exception,
floating_point_underflow_exception): Likewise.
* include/cln/number_io.h (read_number_exception,
read_number_bad_syntax_exception, read_number_junk_exception,
read_number_eof_exception): Likewise.
* include/cln/object.h (CL_DEFINE_CONVERTER): Replace dynamic assertion
with a compile-time assertion.
* include/cln/GV.h: Replace nonreturning functions with exceptions.
* include/cln/modinteger.h: Likewise.
* include/cln/SV.h: Likewise.
* include/cln/ring.h: Likewise.
* include/cln/string.h: Likewise.
* include/cln/univpoly.h: Likewise.
* src/base/cl_abort.cc: Removed.
* src/base/cl_N_err_d0.cc: Removed.
* src/base/cl_d0_exception.cc: New file.
* src/base/cl_as_err.cc: Removed.
* src/base/cl_as_exception.cc: New file.
* src/base/cl_notreached.cc: Removed.
* src/base/cl_notreached_exception.cc: New file.
* src/base/input/cl_read_err_bad.cc: Removed.
* src/base/input/cl_read_bad_syntax_exception.cc: New file.
* src/base/input/cl_read_err_junk.cc: Removed.
* src/base/input/cl_read_junk_exception.cc: New file.
* src/base/input/cl_read_err_eof.cc: Removed.
* src/base/input/cl_read_eof_exception.cc: New file.
* src/base/cl_N.h (cl_as_error): Removed (see cln/exception.h).
* src/base/macros.h (NOTREACHED): Throw.
* src/base/cl_malloc.cc (xmalloc): Throw.
* src/base/digitseq/cl_2DS_div.cc: Throw.
* src/base/digitseq/cl_DS_div.cc: Throw.
* src/base/digitseq/cl_DS_mul.cc: Throw.
* src/base/digitseq/cl_DS_mul_fftc.h: Throw.
* src/base/digitseq/cl_DS_mul_fftcs.h: Throw.
* src/base/digitseq/cl_DS_mul_fftm.h: Throw.
* src/base/digitseq/cl_DS_mul_fftp.h: Throw.
* src/base/digitseq/cl_DS_mul_fftp3.h: Throw.
* src/base/digitseq/cl_DS_mul_fftp3m.h: Throw.
* src/base/digitseq/cl_DS_mul_fftr.h: Throw.
* src/base/digitseq/cl_DS_mul_nuss.h: Throw.
* src/base/digitseq/cl_DS_recipsqrt.cc: Throw.
* src/base/digitseq/cl_DS_sqrt.cc: Throw.
* src/base/hash/cl_hash.h: Throw.
* src/base/hash/cl_hash1.h: Throw.
* src/base/hash/cl_hash1weak.h: Throw.
* src/base/hash/cl_hash2.h: Throw.
* src/base/hash/cl_hash2weak.h: Throw.
* src/base/hash/cl_hashset.h: Throw.
* src/base/hash/cl_hashuniq.h: Throw.
* src/base/hash/cl_hashuniqweak.h: Throw.
* src/base/proplist/cl_pl_add.cc: Throw.
* src/base/ring/cl_no_ring.cc: Throw.
* src/base/string/cl_spushstring.h: Throw.
* src/base/symbol/cl_symbol.cc: Throw.
* src/integer/bitwise/cl_I_ash.h: Removed (see cln/integer.h.)
* src/integer/bitwise/cl_I_asherr.cc: Removed.
* src/integer/bitwise/cl_I_ash_exception.cc: New file.
* src/integer/division/cl_I_exquoerr.cc: Removed.
* src/integer/division/cl_I_exquo_exception.cc: New file.
* src/integer/cl_I.h: Throw.
* src/integer/division/cl_I_exquopos.cc: Throw.
* src/integer/bitwise/cl_I_logbitp_I.cc: Throw.
* src/integer/bitwise/cl_I_ash.cc: Throw.
* src/integer/bitwise/cl_I_ash_I.cc: Throw.
* src/integer/division/cl_I_exquo.cc: Throw.
* src/integer/gcd/cl_I_gcd_aux2.cc: Throw.
* src/integer/conv/cl_I_to_L.cc: Throw.
* src/integer/conv/cl_I_to_Q.cc: Throw.
* src/integer/conv/cl_I_to_UL.cc: Throw.
* src/integer/conv/cl_I_to_UQ.cc: Throw.
* src/integer/conv/cl_I_to_digits.cc: Throw.
* src/integer/elem/cl_I_div.cc: Throw.
* src/integer/algebraic/cl_I_sqrt.cc: Throw.
* src/integer/input/cl_I_read.cc: Throw.
* src/integer/input/cl_I_read_stream.cc: Throw.
* src/integer/misc/cl_I_as.cc: Throw.
* src/rational/elem/cl_RA_from_I_I_div.cc: Throw.
* src/rational/elem/cl_RA_recip.cc: Throw.
* src/rational/input/cl_RA_read.cc: Throw.
* src/rational/input/cl_RA_read_stream.cc: Throw.
* src/rational/input/cl_RA_readparsed.cc: Throw.
* src/rational/misc/cl_RA_as.cc: Throw.
* src/float/base/cl_F_err_nan.cc: Removed.
* src/float/base/cl_F_nan_exception.cc: New file.
* src/float/base/cl_F_err_un.cc: Removed.
* src/float/base/cl_F_underflow_exception.cc: New file.
* src/float/base/cl_F_err_ov.cc: Removed.
* src/float/base/cl_F_overflow_exception.cc: New file.
* src/float/cl_F.h (cl_error_floating_point_nan,
cl_error_floating_point_overflow, cl_error_floating_point_underflow):
Removed (see cln/float.h.)
* src/float/sfloat/cl_SF.h: Throw.
* src/float/sfloat/elem/cl_SF_div.cc: Throw.
* src/float/sfloat/elem/cl_SF_from_RA.cc: Throw.
* src/float/sfloat/elem/cl_SF_scale.cc: Throw.
* src/float/sfloat/elem/cl_SF_scale_I.cc: Throw.
* src/float/sfloat/misc/cl_SF_as.cc: Throw.
* src/float/ffloat/cl_FF.h: Throw.
* src/float/ffloat/conv/cl_FF_from_float.cc: Throw.
* src/float/ffloat/elem/cl_FF_div.cc: Throw.
* src/float/ffloat/elem/cl_FF_from_RA.cc: Throw.
* src/float/ffloat/elem/cl_FF_scale.cc: Throw.
* src/float/ffloat/elem/cl_FF_scale_I.cc: Throw.
* src/float/ffloat/misc/cl_FF_as.cc: Throw.
* src/float/dfloat/cl_DF.h: Throw.
* src/float/dfloat/conv/cl_DF_from_double.cc: Throw.
* src/float/dfloat/elem/cl_DF_div.cc: Throw.
* src/float/dfloat/elem/cl_DF_from_RA.cc: Throw.
* src/float/dfloat/elem/cl_DF_scale.cc: Throw.
* src/float/dfloat/elem/cl_DF_scale_I.cc: Throw.
* src/float/dfloat/misc/cl_DF_as.cc: Throw.
* src/float/lfloat/algebraic/cl_LF_sqrt.cc: Throw.
* src/float/lfloat/elem/cl_LF_1plus.cc: Throw.
* src/float/lfloat/elem/cl_LF_I_div.cc: Throw.
* src/float/lfloat/elem/cl_LF_I_mul.cc: Throw.
* src/float/lfloat/elem/cl_LF_div.cc: Throw.
* src/float/lfloat/elem/cl_LF_from_I.cc: Throw.
* src/float/lfloat/elem/cl_LF_mul.cc: Throw.
* src/float/lfloat/elem/cl_LF_scale.cc: Throw.
* src/float/lfloat/elem/cl_LF_scale_I.cc: Throw.
* src/float/lfloat/elem/cl_LF_square.cc: Throw.
* src/float/lfloat/misc/cl_LF_as.cc: Throw.
* src/float/lfloat/misc/cl_LF_shorten.cc: Throw.
* src/float/lfloat/misc/cl_LF_shortenrel.cc: Throw.
* src/float/lfloat/misc/cl_LF_shortenwith.cc: Throw.
* src/float/input/cl_F_read.cc: Throw.
* src/float/input/cl_F_read_stream.cc: Throw.
* src/float/misc/cl_F_as.cc: Throw.
* src/float/misc/cl_F_shortenrel.cc: Throw.
* src/float/transcendental/cl_LF_coshsinh_aux.cc: Throw.
* src/float/transcendental/cl_LF_cossin_aux.cc: Throw.
* src/float/transcendental/cl_LF_exp_aux.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_a.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_ab.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_b.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_p.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pa.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pab.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pb.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pq.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pqa.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pqab.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pqb.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_q.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_qa.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_qab.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_qb.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_stream_pq.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_stream_pqa.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_stream_pqab.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_stream_pqb.cc: Throw.
* src/float/transcendental/cl_LF_ratsumseries_pqcd_aux.cc: Throw.
* src/float/transcendental/cl_LF_ratsumseries_pqd.cc: Throw.
* src/float/transcendental/cl_LF_ratsumseries_pqd_aux.cc: Throw.
* src/float/transcendental/cl_LF_zeta_int.cc: Throw.
* src/real/elem/cl_R_div.cc: Throw.
* src/real/format-output/cl_fmt_cardinal.cc: Throw.
* src/real/format-output/cl_fmt_newroman.cc: Throw.
* src/real/format-output/cl_fmt_oldroman.cc: Throw.
* src/real/input/cl_R_read.cc: Throw.
* src/real/input/cl_R_read_stream.cc: Throw.
* src/real/misc/cl_R_as.cc: Throw.
* src/real/random/cl_R_random.cc: Throw.
* src/real/transcendental/cl_R_atan2.cc: Throw.
* src/real/transcendental/cl_R_log.cc: Throw.
* src/complex/input/cl_N_read.cc: Throw.
* src/complex/input/cl_N_read_stream.cc: Throw.
* src/complex/misc/cl_N_as.cc: Throw.
* src/complex/transcendental/cl_C_atanh_aux.cc: Throw.
* src/complex/transcendental/cl_C_expt_C.cc: Throw.
* src/complex/transcendental/cl_C_log.cc: Throw.
* src/complex/transcendental/cl_C_log2.cc: Throw.
* src/numtheory/cl_nt_cornacchia1.cc: Throw.
* src/numtheory/cl_nt_cornacchia4.cc: Throw.
* src/numtheory/cl_nt_isprobprime.cc: Throw.
* src/numtheory/cl_nt_jacobi.cc: Throw.
* src/numtheory/cl_nt_jacobi_low.cc: Throw.
* src/numtheory/cl_nt_sqrtmodp.cc: Throw.
* src/modinteger/cl_MI.cc: Throw.
* src/modinteger/cl_MI_int.h: Throw.
* src/modinteger/cl_MI_montgom.h: Throw.
* src/modinteger/cl_MI_pow2.h: Throw.
* src/modinteger/cl_MI_rshift.cc: Throw.
* src/modinteger/cl_MI_std.h: Throw.
* src/polynomial/elem/cl_UP_GF2.h: Throw.
* src/polynomial/elem/cl_UP_MI.h: Throw.
* src/polynomial/elem/cl_UP_gen.h: Throw.
* src/polynomial/elem/cl_UP_named.cc: Throw.
* src/polynomial/elem/cl_UP_no_ring.cc (uninitialized_error,
uninitialized_ring): Removed (see cln/ring.h.)
* src/polynomial/elem/cl_UP_number.h: Throw.
* src/polynomial/elem/cl_UP_unnamed.cc: Throw.
* src/vector/cl_GV_I.cc: Throw.
* src/vector/cl_GV_number.cc: Throw.
* tests/timediv2adic-compare.cc: Use default abort(), not cl_abort().
* tests/timeprint-compare.cc: Likewise.
* tests/timerecip2adic-compare.cc: Likewise.
* doc/cln.tex: Document the exception classes.
* examples/contfrac.cc: Use try/catch instead of setjmp/longjmp.
* INSTALL: undocument -fno-exceptions.
* README: Add exceptions to list of used C++ features.
18 years ago Remove exception hooks in favor of real C++ exceptions:
* include/cln/exception.h: New file...
* include/cln/cln.h: ...included here...
* include/cln/rational.h: ...and here.
* include/cln/abort.h: Removed.
* include/cln/integer.h (ash_exception, exquo_exception): Add exception
types.
* include/cln/float.h (floating_point_exception,
floating_point_nan_exception, floating_point_overflow_exception,
floating_point_underflow_exception): Likewise.
* include/cln/number_io.h (read_number_exception,
read_number_bad_syntax_exception, read_number_junk_exception,
read_number_eof_exception): Likewise.
* include/cln/object.h (CL_DEFINE_CONVERTER): Replace dynamic assertion
with a compile-time assertion.
* include/cln/GV.h: Replace nonreturning functions with exceptions.
* include/cln/modinteger.h: Likewise.
* include/cln/SV.h: Likewise.
* include/cln/ring.h: Likewise.
* include/cln/string.h: Likewise.
* include/cln/univpoly.h: Likewise.
* src/base/cl_abort.cc: Removed.
* src/base/cl_N_err_d0.cc: Removed.
* src/base/cl_d0_exception.cc: New file.
* src/base/cl_as_err.cc: Removed.
* src/base/cl_as_exception.cc: New file.
* src/base/cl_notreached.cc: Removed.
* src/base/cl_notreached_exception.cc: New file.
* src/base/input/cl_read_err_bad.cc: Removed.
* src/base/input/cl_read_bad_syntax_exception.cc: New file.
* src/base/input/cl_read_err_junk.cc: Removed.
* src/base/input/cl_read_junk_exception.cc: New file.
* src/base/input/cl_read_err_eof.cc: Removed.
* src/base/input/cl_read_eof_exception.cc: New file.
* src/base/cl_N.h (cl_as_error): Removed (see cln/exception.h).
* src/base/macros.h (NOTREACHED): Throw.
* src/base/cl_malloc.cc (xmalloc): Throw.
* src/base/digitseq/cl_2DS_div.cc: Throw.
* src/base/digitseq/cl_DS_div.cc: Throw.
* src/base/digitseq/cl_DS_mul.cc: Throw.
* src/base/digitseq/cl_DS_mul_fftc.h: Throw.
* src/base/digitseq/cl_DS_mul_fftcs.h: Throw.
* src/base/digitseq/cl_DS_mul_fftm.h: Throw.
* src/base/digitseq/cl_DS_mul_fftp.h: Throw.
* src/base/digitseq/cl_DS_mul_fftp3.h: Throw.
* src/base/digitseq/cl_DS_mul_fftp3m.h: Throw.
* src/base/digitseq/cl_DS_mul_fftr.h: Throw.
* src/base/digitseq/cl_DS_mul_nuss.h: Throw.
* src/base/digitseq/cl_DS_recipsqrt.cc: Throw.
* src/base/digitseq/cl_DS_sqrt.cc: Throw.
* src/base/hash/cl_hash.h: Throw.
* src/base/hash/cl_hash1.h: Throw.
* src/base/hash/cl_hash1weak.h: Throw.
* src/base/hash/cl_hash2.h: Throw.
* src/base/hash/cl_hash2weak.h: Throw.
* src/base/hash/cl_hashset.h: Throw.
* src/base/hash/cl_hashuniq.h: Throw.
* src/base/hash/cl_hashuniqweak.h: Throw.
* src/base/proplist/cl_pl_add.cc: Throw.
* src/base/ring/cl_no_ring.cc: Throw.
* src/base/string/cl_spushstring.h: Throw.
* src/base/symbol/cl_symbol.cc: Throw.
* src/integer/bitwise/cl_I_ash.h: Removed (see cln/integer.h.)
* src/integer/bitwise/cl_I_asherr.cc: Removed.
* src/integer/bitwise/cl_I_ash_exception.cc: New file.
* src/integer/division/cl_I_exquoerr.cc: Removed.
* src/integer/division/cl_I_exquo_exception.cc: New file.
* src/integer/cl_I.h: Throw.
* src/integer/division/cl_I_exquopos.cc: Throw.
* src/integer/bitwise/cl_I_logbitp_I.cc: Throw.
* src/integer/bitwise/cl_I_ash.cc: Throw.
* src/integer/bitwise/cl_I_ash_I.cc: Throw.
* src/integer/division/cl_I_exquo.cc: Throw.
* src/integer/gcd/cl_I_gcd_aux2.cc: Throw.
* src/integer/conv/cl_I_to_L.cc: Throw.
* src/integer/conv/cl_I_to_Q.cc: Throw.
* src/integer/conv/cl_I_to_UL.cc: Throw.
* src/integer/conv/cl_I_to_UQ.cc: Throw.
* src/integer/conv/cl_I_to_digits.cc: Throw.
* src/integer/elem/cl_I_div.cc: Throw.
* src/integer/algebraic/cl_I_sqrt.cc: Throw.
* src/integer/input/cl_I_read.cc: Throw.
* src/integer/input/cl_I_read_stream.cc: Throw.
* src/integer/misc/cl_I_as.cc: Throw.
* src/rational/elem/cl_RA_from_I_I_div.cc: Throw.
* src/rational/elem/cl_RA_recip.cc: Throw.
* src/rational/input/cl_RA_read.cc: Throw.
* src/rational/input/cl_RA_read_stream.cc: Throw.
* src/rational/input/cl_RA_readparsed.cc: Throw.
* src/rational/misc/cl_RA_as.cc: Throw.
* src/float/base/cl_F_err_nan.cc: Removed.
* src/float/base/cl_F_nan_exception.cc: New file.
* src/float/base/cl_F_err_un.cc: Removed.
* src/float/base/cl_F_underflow_exception.cc: New file.
* src/float/base/cl_F_err_ov.cc: Removed.
* src/float/base/cl_F_overflow_exception.cc: New file.
* src/float/cl_F.h (cl_error_floating_point_nan,
cl_error_floating_point_overflow, cl_error_floating_point_underflow):
Removed (see cln/float.h.)
* src/float/sfloat/cl_SF.h: Throw.
* src/float/sfloat/elem/cl_SF_div.cc: Throw.
* src/float/sfloat/elem/cl_SF_from_RA.cc: Throw.
* src/float/sfloat/elem/cl_SF_scale.cc: Throw.
* src/float/sfloat/elem/cl_SF_scale_I.cc: Throw.
* src/float/sfloat/misc/cl_SF_as.cc: Throw.
* src/float/ffloat/cl_FF.h: Throw.
* src/float/ffloat/conv/cl_FF_from_float.cc: Throw.
* src/float/ffloat/elem/cl_FF_div.cc: Throw.
* src/float/ffloat/elem/cl_FF_from_RA.cc: Throw.
* src/float/ffloat/elem/cl_FF_scale.cc: Throw.
* src/float/ffloat/elem/cl_FF_scale_I.cc: Throw.
* src/float/ffloat/misc/cl_FF_as.cc: Throw.
* src/float/dfloat/cl_DF.h: Throw.
* src/float/dfloat/conv/cl_DF_from_double.cc: Throw.
* src/float/dfloat/elem/cl_DF_div.cc: Throw.
* src/float/dfloat/elem/cl_DF_from_RA.cc: Throw.
* src/float/dfloat/elem/cl_DF_scale.cc: Throw.
* src/float/dfloat/elem/cl_DF_scale_I.cc: Throw.
* src/float/dfloat/misc/cl_DF_as.cc: Throw.
* src/float/lfloat/algebraic/cl_LF_sqrt.cc: Throw.
* src/float/lfloat/elem/cl_LF_1plus.cc: Throw.
* src/float/lfloat/elem/cl_LF_I_div.cc: Throw.
* src/float/lfloat/elem/cl_LF_I_mul.cc: Throw.
* src/float/lfloat/elem/cl_LF_div.cc: Throw.
* src/float/lfloat/elem/cl_LF_from_I.cc: Throw.
* src/float/lfloat/elem/cl_LF_mul.cc: Throw.
* src/float/lfloat/elem/cl_LF_scale.cc: Throw.
* src/float/lfloat/elem/cl_LF_scale_I.cc: Throw.
* src/float/lfloat/elem/cl_LF_square.cc: Throw.
* src/float/lfloat/misc/cl_LF_as.cc: Throw.
* src/float/lfloat/misc/cl_LF_shorten.cc: Throw.
* src/float/lfloat/misc/cl_LF_shortenrel.cc: Throw.
* src/float/lfloat/misc/cl_LF_shortenwith.cc: Throw.
* src/float/input/cl_F_read.cc: Throw.
* src/float/input/cl_F_read_stream.cc: Throw.
* src/float/misc/cl_F_as.cc: Throw.
* src/float/misc/cl_F_shortenrel.cc: Throw.
* src/float/transcendental/cl_LF_coshsinh_aux.cc: Throw.
* src/float/transcendental/cl_LF_cossin_aux.cc: Throw.
* src/float/transcendental/cl_LF_exp_aux.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_a.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_ab.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_b.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_p.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pa.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pab.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pb.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pq.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pqa.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pqab.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pqb.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_q.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_qa.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_qab.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_qb.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_stream_pq.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_stream_pqa.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_stream_pqab.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_stream_pqb.cc: Throw.
* src/float/transcendental/cl_LF_ratsumseries_pqcd_aux.cc: Throw.
* src/float/transcendental/cl_LF_ratsumseries_pqd.cc: Throw.
* src/float/transcendental/cl_LF_ratsumseries_pqd_aux.cc: Throw.
* src/float/transcendental/cl_LF_zeta_int.cc: Throw.
* src/real/elem/cl_R_div.cc: Throw.
* src/real/format-output/cl_fmt_cardinal.cc: Throw.
* src/real/format-output/cl_fmt_newroman.cc: Throw.
* src/real/format-output/cl_fmt_oldroman.cc: Throw.
* src/real/input/cl_R_read.cc: Throw.
* src/real/input/cl_R_read_stream.cc: Throw.
* src/real/misc/cl_R_as.cc: Throw.
* src/real/random/cl_R_random.cc: Throw.
* src/real/transcendental/cl_R_atan2.cc: Throw.
* src/real/transcendental/cl_R_log.cc: Throw.
* src/complex/input/cl_N_read.cc: Throw.
* src/complex/input/cl_N_read_stream.cc: Throw.
* src/complex/misc/cl_N_as.cc: Throw.
* src/complex/transcendental/cl_C_atanh_aux.cc: Throw.
* src/complex/transcendental/cl_C_expt_C.cc: Throw.
* src/complex/transcendental/cl_C_log.cc: Throw.
* src/complex/transcendental/cl_C_log2.cc: Throw.
* src/numtheory/cl_nt_cornacchia1.cc: Throw.
* src/numtheory/cl_nt_cornacchia4.cc: Throw.
* src/numtheory/cl_nt_isprobprime.cc: Throw.
* src/numtheory/cl_nt_jacobi.cc: Throw.
* src/numtheory/cl_nt_jacobi_low.cc: Throw.
* src/numtheory/cl_nt_sqrtmodp.cc: Throw.
* src/modinteger/cl_MI.cc: Throw.
* src/modinteger/cl_MI_int.h: Throw.
* src/modinteger/cl_MI_montgom.h: Throw.
* src/modinteger/cl_MI_pow2.h: Throw.
* src/modinteger/cl_MI_rshift.cc: Throw.
* src/modinteger/cl_MI_std.h: Throw.
* src/polynomial/elem/cl_UP_GF2.h: Throw.
* src/polynomial/elem/cl_UP_MI.h: Throw.
* src/polynomial/elem/cl_UP_gen.h: Throw.
* src/polynomial/elem/cl_UP_named.cc: Throw.
* src/polynomial/elem/cl_UP_no_ring.cc (uninitialized_error,
uninitialized_ring): Removed (see cln/ring.h.)
* src/polynomial/elem/cl_UP_number.h: Throw.
* src/polynomial/elem/cl_UP_unnamed.cc: Throw.
* src/vector/cl_GV_I.cc: Throw.
* src/vector/cl_GV_number.cc: Throw.
* tests/timediv2adic-compare.cc: Use default abort(), not cl_abort().
* tests/timeprint-compare.cc: Likewise.
* tests/timerecip2adic-compare.cc: Likewise.
* doc/cln.tex: Document the exception classes.
* examples/contfrac.cc: Use try/catch instead of setjmp/longjmp.
* INSTALL: undocument -fno-exceptions.
* README: Add exceptions to list of used C++ features.
18 years ago Remove exception hooks in favor of real C++ exceptions:
* include/cln/exception.h: New file...
* include/cln/cln.h: ...included here...
* include/cln/rational.h: ...and here.
* include/cln/abort.h: Removed.
* include/cln/integer.h (ash_exception, exquo_exception): Add exception
types.
* include/cln/float.h (floating_point_exception,
floating_point_nan_exception, floating_point_overflow_exception,
floating_point_underflow_exception): Likewise.
* include/cln/number_io.h (read_number_exception,
read_number_bad_syntax_exception, read_number_junk_exception,
read_number_eof_exception): Likewise.
* include/cln/object.h (CL_DEFINE_CONVERTER): Replace dynamic assertion
with a compile-time assertion.
* include/cln/GV.h: Replace nonreturning functions with exceptions.
* include/cln/modinteger.h: Likewise.
* include/cln/SV.h: Likewise.
* include/cln/ring.h: Likewise.
* include/cln/string.h: Likewise.
* include/cln/univpoly.h: Likewise.
* src/base/cl_abort.cc: Removed.
* src/base/cl_N_err_d0.cc: Removed.
* src/base/cl_d0_exception.cc: New file.
* src/base/cl_as_err.cc: Removed.
* src/base/cl_as_exception.cc: New file.
* src/base/cl_notreached.cc: Removed.
* src/base/cl_notreached_exception.cc: New file.
* src/base/input/cl_read_err_bad.cc: Removed.
* src/base/input/cl_read_bad_syntax_exception.cc: New file.
* src/base/input/cl_read_err_junk.cc: Removed.
* src/base/input/cl_read_junk_exception.cc: New file.
* src/base/input/cl_read_err_eof.cc: Removed.
* src/base/input/cl_read_eof_exception.cc: New file.
* src/base/cl_N.h (cl_as_error): Removed (see cln/exception.h).
* src/base/macros.h (NOTREACHED): Throw.
* src/base/cl_malloc.cc (xmalloc): Throw.
* src/base/digitseq/cl_2DS_div.cc: Throw.
* src/base/digitseq/cl_DS_div.cc: Throw.
* src/base/digitseq/cl_DS_mul.cc: Throw.
* src/base/digitseq/cl_DS_mul_fftc.h: Throw.
* src/base/digitseq/cl_DS_mul_fftcs.h: Throw.
* src/base/digitseq/cl_DS_mul_fftm.h: Throw.
* src/base/digitseq/cl_DS_mul_fftp.h: Throw.
* src/base/digitseq/cl_DS_mul_fftp3.h: Throw.
* src/base/digitseq/cl_DS_mul_fftp3m.h: Throw.
* src/base/digitseq/cl_DS_mul_fftr.h: Throw.
* src/base/digitseq/cl_DS_mul_nuss.h: Throw.
* src/base/digitseq/cl_DS_recipsqrt.cc: Throw.
* src/base/digitseq/cl_DS_sqrt.cc: Throw.
* src/base/hash/cl_hash.h: Throw.
* src/base/hash/cl_hash1.h: Throw.
* src/base/hash/cl_hash1weak.h: Throw.
* src/base/hash/cl_hash2.h: Throw.
* src/base/hash/cl_hash2weak.h: Throw.
* src/base/hash/cl_hashset.h: Throw.
* src/base/hash/cl_hashuniq.h: Throw.
* src/base/hash/cl_hashuniqweak.h: Throw.
* src/base/proplist/cl_pl_add.cc: Throw.
* src/base/ring/cl_no_ring.cc: Throw.
* src/base/string/cl_spushstring.h: Throw.
* src/base/symbol/cl_symbol.cc: Throw.
* src/integer/bitwise/cl_I_ash.h: Removed (see cln/integer.h.)
* src/integer/bitwise/cl_I_asherr.cc: Removed.
* src/integer/bitwise/cl_I_ash_exception.cc: New file.
* src/integer/division/cl_I_exquoerr.cc: Removed.
* src/integer/division/cl_I_exquo_exception.cc: New file.
* src/integer/cl_I.h: Throw.
* src/integer/division/cl_I_exquopos.cc: Throw.
* src/integer/bitwise/cl_I_logbitp_I.cc: Throw.
* src/integer/bitwise/cl_I_ash.cc: Throw.
* src/integer/bitwise/cl_I_ash_I.cc: Throw.
* src/integer/division/cl_I_exquo.cc: Throw.
* src/integer/gcd/cl_I_gcd_aux2.cc: Throw.
* src/integer/conv/cl_I_to_L.cc: Throw.
* src/integer/conv/cl_I_to_Q.cc: Throw.
* src/integer/conv/cl_I_to_UL.cc: Throw.
* src/integer/conv/cl_I_to_UQ.cc: Throw.
* src/integer/conv/cl_I_to_digits.cc: Throw.
* src/integer/elem/cl_I_div.cc: Throw.
* src/integer/algebraic/cl_I_sqrt.cc: Throw.
* src/integer/input/cl_I_read.cc: Throw.
* src/integer/input/cl_I_read_stream.cc: Throw.
* src/integer/misc/cl_I_as.cc: Throw.
* src/rational/elem/cl_RA_from_I_I_div.cc: Throw.
* src/rational/elem/cl_RA_recip.cc: Throw.
* src/rational/input/cl_RA_read.cc: Throw.
* src/rational/input/cl_RA_read_stream.cc: Throw.
* src/rational/input/cl_RA_readparsed.cc: Throw.
* src/rational/misc/cl_RA_as.cc: Throw.
* src/float/base/cl_F_err_nan.cc: Removed.
* src/float/base/cl_F_nan_exception.cc: New file.
* src/float/base/cl_F_err_un.cc: Removed.
* src/float/base/cl_F_underflow_exception.cc: New file.
* src/float/base/cl_F_err_ov.cc: Removed.
* src/float/base/cl_F_overflow_exception.cc: New file.
* src/float/cl_F.h (cl_error_floating_point_nan,
cl_error_floating_point_overflow, cl_error_floating_point_underflow):
Removed (see cln/float.h.)
* src/float/sfloat/cl_SF.h: Throw.
* src/float/sfloat/elem/cl_SF_div.cc: Throw.
* src/float/sfloat/elem/cl_SF_from_RA.cc: Throw.
* src/float/sfloat/elem/cl_SF_scale.cc: Throw.
* src/float/sfloat/elem/cl_SF_scale_I.cc: Throw.
* src/float/sfloat/misc/cl_SF_as.cc: Throw.
* src/float/ffloat/cl_FF.h: Throw.
* src/float/ffloat/conv/cl_FF_from_float.cc: Throw.
* src/float/ffloat/elem/cl_FF_div.cc: Throw.
* src/float/ffloat/elem/cl_FF_from_RA.cc: Throw.
* src/float/ffloat/elem/cl_FF_scale.cc: Throw.
* src/float/ffloat/elem/cl_FF_scale_I.cc: Throw.
* src/float/ffloat/misc/cl_FF_as.cc: Throw.
* src/float/dfloat/cl_DF.h: Throw.
* src/float/dfloat/conv/cl_DF_from_double.cc: Throw.
* src/float/dfloat/elem/cl_DF_div.cc: Throw.
* src/float/dfloat/elem/cl_DF_from_RA.cc: Throw.
* src/float/dfloat/elem/cl_DF_scale.cc: Throw.
* src/float/dfloat/elem/cl_DF_scale_I.cc: Throw.
* src/float/dfloat/misc/cl_DF_as.cc: Throw.
* src/float/lfloat/algebraic/cl_LF_sqrt.cc: Throw.
* src/float/lfloat/elem/cl_LF_1plus.cc: Throw.
* src/float/lfloat/elem/cl_LF_I_div.cc: Throw.
* src/float/lfloat/elem/cl_LF_I_mul.cc: Throw.
* src/float/lfloat/elem/cl_LF_div.cc: Throw.
* src/float/lfloat/elem/cl_LF_from_I.cc: Throw.
* src/float/lfloat/elem/cl_LF_mul.cc: Throw.
* src/float/lfloat/elem/cl_LF_scale.cc: Throw.
* src/float/lfloat/elem/cl_LF_scale_I.cc: Throw.
* src/float/lfloat/elem/cl_LF_square.cc: Throw.
* src/float/lfloat/misc/cl_LF_as.cc: Throw.
* src/float/lfloat/misc/cl_LF_shorten.cc: Throw.
* src/float/lfloat/misc/cl_LF_shortenrel.cc: Throw.
* src/float/lfloat/misc/cl_LF_shortenwith.cc: Throw.
* src/float/input/cl_F_read.cc: Throw.
* src/float/input/cl_F_read_stream.cc: Throw.
* src/float/misc/cl_F_as.cc: Throw.
* src/float/misc/cl_F_shortenrel.cc: Throw.
* src/float/transcendental/cl_LF_coshsinh_aux.cc: Throw.
* src/float/transcendental/cl_LF_cossin_aux.cc: Throw.
* src/float/transcendental/cl_LF_exp_aux.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_a.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_ab.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_b.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_p.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pa.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pab.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pb.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pq.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pqa.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pqab.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_pqb.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_q.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_qa.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_qab.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_qb.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_stream_pq.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_stream_pqa.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_stream_pqab.cc: Throw.
* src/float/transcendental/cl_LF_ratseries_stream_pqb.cc: Throw.
* src/float/transcendental/cl_LF_ratsumseries_pqcd_aux.cc: Throw.
* src/float/transcendental/cl_LF_ratsumseries_pqd.cc: Throw.
* src/float/transcendental/cl_LF_ratsumseries_pqd_aux.cc: Throw.
* src/float/transcendental/cl_LF_zeta_int.cc: Throw.
* src/real/elem/cl_R_div.cc: Throw.
* src/real/format-output/cl_fmt_cardinal.cc: Throw.
* src/real/format-output/cl_fmt_newroman.cc: Throw.
* src/real/format-output/cl_fmt_oldroman.cc: Throw.
* src/real/input/cl_R_read.cc: Throw.
* src/real/input/cl_R_read_stream.cc: Throw.
* src/real/misc/cl_R_as.cc: Throw.
* src/real/random/cl_R_random.cc: Throw.
* src/real/transcendental/cl_R_atan2.cc: Throw.
* src/real/transcendental/cl_R_log.cc: Throw.
* src/complex/input/cl_N_read.cc: Throw.
* src/complex/input/cl_N_read_stream.cc: Throw.
* src/complex/misc/cl_N_as.cc: Throw.
* src/complex/transcendental/cl_C_atanh_aux.cc: Throw.
* src/complex/transcendental/cl_C_expt_C.cc: Throw.
* src/complex/transcendental/cl_C_log.cc: Throw.
* src/complex/transcendental/cl_C_log2.cc: Throw.
* src/numtheory/cl_nt_cornacchia1.cc: Throw.
* src/numtheory/cl_nt_cornacchia4.cc: Throw.
* src/numtheory/cl_nt_isprobprime.cc: Throw.
* src/numtheory/cl_nt_jacobi.cc: Throw.
* src/numtheory/cl_nt_jacobi_low.cc: Throw.
* src/numtheory/cl_nt_sqrtmodp.cc: Throw.
* src/modinteger/cl_MI.cc: Throw.
* src/modinteger/cl_MI_int.h: Throw.
* src/modinteger/cl_MI_montgom.h: Throw.
* src/modinteger/cl_MI_pow2.h: Throw.
* src/modinteger/cl_MI_rshift.cc: Throw.
* src/modinteger/cl_MI_std.h: Throw.
* src/polynomial/elem/cl_UP_GF2.h: Throw.
* src/polynomial/elem/cl_UP_MI.h: Throw.
* src/polynomial/elem/cl_UP_gen.h: Throw.
* src/polynomial/elem/cl_UP_named.cc: Throw.
* src/polynomial/elem/cl_UP_no_ring.cc (uninitialized_error,
uninitialized_ring): Removed (see cln/ring.h.)
* src/polynomial/elem/cl_UP_number.h: Throw.
* src/polynomial/elem/cl_UP_unnamed.cc: Throw.
* src/vector/cl_GV_I.cc: Throw.
* src/vector/cl_GV_number.cc: Throw.
* tests/timediv2adic-compare.cc: Use default abort(), not cl_abort().
* tests/timeprint-compare.cc: Likewise.
* tests/timerecip2adic-compare.cc: Likewise.
* doc/cln.tex: Document the exception classes.
* examples/contfrac.cc: Use try/catch instead of setjmp/longjmp.
* INSTALL: undocument -fno-exceptions.
* README: Add exceptions to list of used C++ features.
18 years ago |
|
// Public float operations.
#ifndef _CL_FLOAT_H
#define _CL_FLOAT_H
#include "cln/number.h"
#include "cln/float_class.h"
#include "cln/floatformat.h"
#include "cln/random.h"
#include "cln/integer_class.h"
#include "cln/sfloat_class.h"
#include "cln/ffloat_class.h"
#include "cln/dfloat_class.h"
#include "cln/lfloat_class.h"
#include "cln/exception.h"
namespace cln {
CL_DEFINE_AS_CONVERSION(cl_F)
// Return type for integer_decode_float:
struct cl_idecoded_float { cl_I mantissa; cl_I exponent; cl_I sign; // Constructor.
cl_idecoded_float () {} cl_idecoded_float (const cl_I& m, const cl_I& e, const cl_I& s) : mantissa(m), exponent(e), sign(s) {} };
// zerop(x) testet, ob (= x 0).
extern bool zerop (const cl_F& x);
// minusp(x) testet, ob (< x 0).
extern bool minusp (const cl_F& x);
// plusp(x) testet, ob (> x 0).
extern bool plusp (const cl_F& x);
// cl_F_to_SF(x) wandelt ein Float x in ein Short-Float um und rundet dabei.
extern const cl_SF cl_F_to_SF (const cl_F& x);
// cl_F_to_FF(x) wandelt ein Float x in ein Single-Float um und rundet dabei.
extern const cl_FF cl_F_to_FF (const cl_F& x);
// cl_F_to_DF(x) wandelt ein Float x in ein Double-Float um und rundet dabei.
extern const cl_DF cl_F_to_DF (const cl_F& x);
// cl_F_to_LF(x,len) wandelt ein Float x in ein Long-Float mit len Digits um
// und rundet dabei.
// > uintC len: gewünschte Anzahl Digits, >=LF_minlen
extern const cl_LF cl_F_to_LF (const cl_F& x, uintC len);
// The default float format used when converting rational numbers to floats.
extern float_format_t default_float_format;
// Returns the smallest float format which guarantees at least n decimal digits
// in the mantissa (after the decimal point).
extern float_format_t float_format (uintE n);
// cl_float(x,y) wandelt ein Float x in das Float-Format des Floats y um
// und rundet dabei nötigenfalls.
// > x,y: Floats
// < ergebnis: (float x y)
extern const cl_F cl_float (const cl_F& x, const cl_F& y);
// cl_float(x,f) wandelt ein Float x in das Float-Format f um
// und rundet dabei nötigenfalls.
// > x: ein Float
// > f: eine Float-Format-Spezifikation
// < ergebnis: (float x f)
extern const cl_F cl_float (const cl_F& x, float_format_t f);
// cl_float(x) wandelt eine reelle Zahl x in ein Float um
// und rundet dabei nötigenfalls.
// > x: eine reelle Zahl
// < ergebnis: (float x)
// Abhängig von default_float_format.
inline const cl_F cl_float (const cl_F& x) { return x; }
// cl_float(x,y) wandelt ein Integer x in das Float-Format des Floats y um
// und rundet dabei nötigenfalls.
// > x: ein Integer
// > y: ein Float
// < ergebnis: (float x y)
extern const cl_F cl_float (const cl_I& x, const cl_F& y);
// cl_float(x,y) wandelt ein Integer x in das Float-Format f um
// und rundet dabei nötigenfalls.
// > x: ein Integer
// > f: eine Float-Format-Spezifikation
// < ergebnis: (float x f)
extern const cl_F cl_float (const cl_I& x, float_format_t f);
// cl_float(x) wandelt ein Integer x in ein Float um und rundet dabei.
// > x: ein Integer
// < ergebnis: (float x)
// Abhängig von default_float_format.
extern const cl_F cl_float (const cl_I& x);
// cl_float(x,y) wandelt eine rationale Zahl x in das Float-Format des
// Floats y um und rundet dabei nötigenfalls.
// > x: eine rationale Zahl
// > y: ein Float
// < ergebnis: (float x y)
extern const cl_F cl_float (const cl_RA& x, const cl_F& y);
// cl_float(x,y) wandelt eine rationale Zahl x in das Float-Format f um
// und rundet dabei nötigenfalls.
// > x: eine rationale Zahl
// > f: eine Float-Format-Spezifikation
// < ergebnis: (float x f)
extern const cl_F cl_float (const cl_RA& x, float_format_t f);
// cl_float(x) wandelt eine rationale Zahl x in ein Float um und rundet dabei.
// > x: eine rationale Zahl
// < ergebnis: (float x)
// Abhängig von default_float_format.
extern const cl_F cl_float (const cl_RA& x);
// The C++ compilers are not clever enough to guess this:
inline const cl_F cl_float (int x, const cl_F& y) { return cl_float(cl_I(x),y); } inline const cl_F cl_float (unsigned int x, const cl_F& y) { return cl_float(cl_I(x),y); } inline const cl_F cl_float (int x, float_format_t y) { return cl_float(cl_I(x),y); } inline const cl_F cl_float (unsigned int x, float_format_t y) { return cl_float(cl_I(x),y); } inline const cl_F cl_float (int x) { return cl_float(cl_I(x)); } inline const cl_F cl_float (unsigned int x) { return cl_float(cl_I(x)); } // The C++ compilers could hardly guess the following:
inline const cl_F cl_float (float x, const cl_F& y) { return cl_float(cl_FF(x),y); } inline const cl_F cl_float (double x, const cl_F& y) { return cl_float(cl_DF(x),y); } inline const cl_F cl_float (float x, float_format_t y) { return cl_float(cl_FF(x),y); } inline const cl_F cl_float (double x, float_format_t y) { return cl_float(cl_DF(x),y); } inline const cl_F cl_float (float x) { return cl_float(cl_FF(x)); } inline const cl_F cl_float (double x) { return cl_float(cl_DF(x)); }
// Liefert (- x), wo x ein Float ist.
extern const cl_F operator- (const cl_F& x);
// Liefert (+ x y), wo x und y Floats sind.
extern const cl_F operator+ (const cl_F& x, const cl_F& y); // The C++ compilers could hardly guess the following:
inline const cl_F operator+ (const cl_RA& x, const cl_F& y) { return cl_float(x,y) + y; } inline const cl_F operator+ (const cl_I& x, const cl_F& y) { return cl_float(x,y) + y; } inline const cl_F operator+ (const cl_F& x, const cl_RA& y) { return x + cl_float(y,x); } inline const cl_F operator+ (const cl_F& x, const cl_I& y) { return x + cl_float(y,x); } // Dem C++-Compiler muß man nun auch das Folgende sagen:
inline const cl_F operator+ (const int x, const cl_F& y) { return cl_I(x) + y; } inline const cl_F operator+ (const unsigned int x, const cl_F& y) { return cl_I(x) + y; } inline const cl_F operator+ (const long x, const cl_F& y) { return cl_I(x) + y; } inline const cl_F operator+ (const unsigned long x, const cl_F& y) { return cl_I(x) + y; } #ifdef HAVE_LONGLONG
inline const cl_F operator+ (const long long x, const cl_F& y) { return cl_I(x) + y; } inline const cl_F operator+ (const unsigned long long x, const cl_F& y) { return cl_I(x) + y; } #endif
inline const cl_F operator+ (const float x, const cl_F& y) { return cl_F(x) + y; } inline const cl_F operator+ (const double x, const cl_F& y) { return cl_F(x) + y; } inline const cl_F operator+ (const cl_F& x, const int y) { return x + cl_I(y); } inline const cl_F operator+ (const cl_F& x, const unsigned int y) { return x + cl_I(y); } inline const cl_F operator+ (const cl_F& x, const long y) { return x + cl_I(y); } inline const cl_F operator+ (const cl_F& x, const unsigned long y) { return x + cl_I(y); } #ifdef HAVE_LONGLONG
inline const cl_F operator+ (const cl_F& x, const long long y) { return x + cl_I(y); } inline const cl_F operator+ (const cl_F& x, const unsigned long long y) { return x + cl_I(y); } #endif
inline const cl_F operator+ (const cl_F& x, const float y) { return x + cl_F(y); } inline const cl_F operator+ (const cl_F& x, const double y) { return x + cl_F(y); }
// Liefert (- x y), wo x und y Floats sind.
extern const cl_F operator- (const cl_F& x, const cl_F& y); // The C++ compilers could hardly guess the following:
inline const cl_F operator- (const cl_RA& x, const cl_F& y) { return cl_float(x,y) - y; } inline const cl_F operator- (const cl_I& x, const cl_F& y) { return cl_float(x,y) - y; } inline const cl_F operator- (const cl_F& x, const cl_RA& y) { return x - cl_float(y,x); } inline const cl_F operator- (const cl_F& x, const cl_I& y) { return x - cl_float(y,x); } // Dem C++-Compiler muß man nun auch das Folgende sagen:
inline const cl_F operator- (const int x, const cl_F& y) { return cl_I(x) - y; } inline const cl_F operator- (const unsigned int x, const cl_F& y) { return cl_I(x) - y; } inline const cl_F operator- (const long x, const cl_F& y) { return cl_I(x) - y; } inline const cl_F operator- (const unsigned long x, const cl_F& y) { return cl_I(x) - y; } #ifdef HAVE_LONGLONG
inline const cl_F operator- (const long long x, const cl_F& y) { return cl_I(x) - y; } inline const cl_F operator- (const unsigned long long x, const cl_F& y) { return cl_I(x) - y; } #endif
inline const cl_F operator- (const float x, const cl_F& y) { return cl_F(x) - y; } inline const cl_F operator- (const double x, const cl_F& y) { return cl_F(x) - y; } inline const cl_F operator- (const cl_F& x, const int y) { return x - cl_I(y); } inline const cl_F operator- (const cl_F& x, const unsigned int y) { return x - cl_I(y); } inline const cl_F operator- (const cl_F& x, const long y) { return x - cl_I(y); } inline const cl_F operator- (const cl_F& x, const unsigned long y) { return x - cl_I(y); } #ifdef HAVE_LONGLONG
inline const cl_F operator- (const cl_F& x, const long long y) { return x - cl_I(y); } inline const cl_F operator- (const cl_F& x, const unsigned long long y) { return x - cl_I(y); } #endif
inline const cl_F operator- (const cl_F& x, const float y) { return x - cl_F(y); } inline const cl_F operator- (const cl_F& x, const double y) { return x - cl_F(y); }
// Liefert (* x y), wo x und y Floats sind.
extern const cl_F operator* (const cl_F& x, const cl_F& y); // Spezialfall x oder y Integer oder rationale Zahl.
inline const cl_R operator* (const cl_F& x, const cl_I& y) { extern const cl_R cl_F_I_mul (const cl_F&, const cl_I&); return cl_F_I_mul(x,y); } inline const cl_R operator* (const cl_I& x, const cl_F& y) { extern const cl_R cl_F_I_mul (const cl_F&, const cl_I&); return cl_F_I_mul(y,x); } inline const cl_R operator* (const cl_F& x, const cl_RA& y) { extern const cl_R cl_F_RA_mul (const cl_F&, const cl_RA&); return cl_F_RA_mul(x,y); } inline const cl_R operator* (const cl_RA& x, const cl_F& y) { extern const cl_R cl_F_RA_mul (const cl_F&, const cl_RA&); return cl_F_RA_mul(y,x); } // Dem C++-Compiler muß man nun auch das Folgende sagen:
inline const cl_R operator* (const int x, const cl_F& y) { return cl_I(x) * y; } inline const cl_R operator* (const unsigned int x, const cl_F& y) { return cl_I(x) * y; } inline const cl_R operator* (const long x, const cl_F& y) { return cl_I(x) * y; } inline const cl_R operator* (const unsigned long x, const cl_F& y) { return cl_I(x) * y; } #ifdef HAVE_LONGLONG
inline const cl_R operator* (const long long x, const cl_F& y) { return cl_I(x) * y; } inline const cl_R operator* (const unsigned long long x, const cl_F& y) { return cl_I(x) * y; } #endif
inline const cl_F operator* (const float x, const cl_F& y) { return cl_F(x) * y; } inline const cl_F operator* (const double x, const cl_F& y) { return cl_F(x) * y; } inline const cl_R operator* (const cl_F& x, const int y) { return x * cl_I(y); } inline const cl_R operator* (const cl_F& x, const unsigned int y) { return x * cl_I(y); } inline const cl_R operator* (const cl_F& x, const long y) { return x * cl_I(y); } inline const cl_R operator* (const cl_F& x, const unsigned long y) { return x * cl_I(y); } #ifdef HAVE_LONGLONG
inline const cl_R operator* (const cl_F& x, const long long y) { return x * cl_I(y); } inline const cl_R operator* (const cl_F& x, const unsigned long long y) { return x * cl_I(y); } #endif
inline const cl_F operator* (const cl_F& x, const float y) { return x * cl_F(y); } inline const cl_F operator* (const cl_F& x, const double y) { return x * cl_F(y); }
// Liefert (* x x), wo x ein Float ist.
extern const cl_F square (const cl_F& x);
// Liefert (/ x y), wo x und y Floats sind.
extern const cl_F operator/ (const cl_F& x, const cl_F& y); // Liefert (/ x y), wo x und y ein Float und eine rationale Zahl sind.
extern const cl_F operator/ (const cl_F& x, const cl_RA& y); extern const cl_F operator/ (const cl_F& x, const cl_I& y); extern const cl_R operator/ (const cl_RA& x, const cl_F& y); extern const cl_R operator/ (const cl_I& x, const cl_F& y); // The C++ compilers could hardly guess the following:
inline const cl_F operator/ (const cl_F& x, const int y) { return x / cl_I(y); } inline const cl_F operator/ (const cl_F& x, const unsigned int y) { return x / cl_I(y); } inline const cl_F operator/ (const cl_F& x, const long y) { return x / cl_I(y); } inline const cl_F operator/ (const cl_F& x, const unsigned long y) { return x / cl_I(y); } #ifdef HAVE_LONGLONG
inline const cl_F operator/ (const cl_F& x, const long long y) { return x / cl_I(y); } inline const cl_F operator/ (const cl_F& x, const unsigned long long y) { return x / cl_I(y); } #endif
inline const cl_F operator/ (const cl_F& x, const float y) { return x / cl_F(y); } inline const cl_F operator/ (const cl_F& x, const double y) { return x / cl_F(y); } inline const cl_R operator/ (const int x, const cl_F& y) { return cl_I(x) / y; } inline const cl_R operator/ (const unsigned int x, const cl_F& y) { return cl_I(x) / y; } inline const cl_R operator/ (const long x, const cl_F& y) { return cl_I(x) / y; } inline const cl_R operator/ (const unsigned long x, const cl_F& y) { return cl_I(x) / y; } #ifdef HAVE_LONGLONG
inline const cl_R operator/ (const long long x, const cl_F& y) { return cl_I(x) / y; } inline const cl_R operator/ (const unsigned long long x, const cl_F& y) { return cl_I(x) / y; } #endif
inline const cl_F operator/ (const float x, const cl_F& y) { return cl_F(x) / y; } inline const cl_F operator/ (const double x, const cl_F& y) { return cl_F(x) / y; }
// Liefert (abs x), wo x ein Float ist.
extern const cl_F abs (const cl_F& x);
// Liefert zu einem Float x>=0 : (sqrt x), ein Float.
extern const cl_F sqrt (const cl_F& x);
// recip(x) liefert (/ x), wo x ein Float ist.
extern const cl_F recip (const cl_F& x);
// (1+ x), wo x ein Float ist.
inline const cl_F plus1 (const cl_F& x) // { return x + cl_I(1); } { return x + cl_float(1,x); }
// (1- x), wo x ein Float ist.
inline const cl_F minus1 (const cl_F& x) // { return x + cl_I(-1); } { return x + cl_float(-1,x); }
// compare(x,y) vergleicht zwei Floats x und y.
// Ergebnis: 0 falls x=y, +1 falls x>y, -1 falls x<y.
extern cl_signean compare (const cl_F& x, const cl_F& y);
// equal_hashcode(x) liefert einen equal-invarianten Hashcode für x.
extern uint32 equal_hashcode (const cl_F& x);
inline bool operator== (const cl_F& x, const cl_F& y) { return compare(x,y)==0; } inline bool operator!= (const cl_F& x, const cl_F& y) { return compare(x,y)!=0; } inline bool operator<= (const cl_F& x, const cl_F& y) { return compare(x,y)<=0; } inline bool operator< (const cl_F& x, const cl_F& y) { return compare(x,y)<0; } inline bool operator>= (const cl_F& x, const cl_F& y) { return compare(x,y)>=0; } inline bool operator> (const cl_F& x, const cl_F& y) { return compare(x,y)>0; }
// ffloor(x) liefert (ffloor x), wo x ein Float ist.
extern const cl_F ffloor (const cl_F& x);
// fceiling(x) liefert (fceiling x), wo x ein Float ist.
extern const cl_F fceiling (const cl_F& x);
// ftruncate(x) liefert (ftruncate x), wo x ein Float ist.
extern const cl_F ftruncate (const cl_F& x);
// fround(x) liefert (fround x), wo x ein Float ist.
extern const cl_F fround (const cl_F& x);
// Return type for frounding operators.
// x / y --> (q,r) with x = y*q+r.
struct cl_F_fdiv_t { cl_F quotient; cl_F remainder; // Constructor.
cl_F_fdiv_t () {} cl_F_fdiv_t (const cl_F& q, const cl_F& r) : quotient(q), remainder(r) {} };
// ffloor2(x) liefert (ffloor x), wo x ein F ist.
extern const cl_F_fdiv_t ffloor2 (const cl_F& x);
// fceiling2(x) liefert (fceiling x), wo x ein F ist.
extern const cl_F_fdiv_t fceiling2 (const cl_F& x);
// ftruncate2(x) liefert (ftruncate x), wo x ein F ist.
extern const cl_F_fdiv_t ftruncate2 (const cl_F& x);
// fround2(x) liefert (fround x), wo x ein F ist.
extern const cl_F_fdiv_t fround2 (const cl_F& x);
// Return type for rounding operators.
// x / y --> (q,r) with x = y*q+r.
struct cl_F_div_t { cl_I quotient; cl_F remainder; // Constructor.
cl_F_div_t () {} cl_F_div_t (const cl_I& q, const cl_F& r) : quotient(q), remainder(r) {} };
// floor2(x) liefert (floor x), wo x ein F ist.
extern const cl_F_div_t floor2 (const cl_F& x); extern const cl_I floor1 (const cl_F& x);
// ceiling2(x) liefert (ceiling x), wo x ein F ist.
extern const cl_F_div_t ceiling2 (const cl_F& x); extern const cl_I ceiling1 (const cl_F& x);
// truncate2(x) liefert (truncate x), wo x ein F ist.
extern const cl_F_div_t truncate2 (const cl_F& x); extern const cl_I truncate1 (const cl_F& x);
// round2(x) liefert (round x), wo x ein F ist.
extern const cl_F_div_t round2 (const cl_F& x); extern const cl_I round1 (const cl_F& x);
// floor2(x,y) liefert (floor x y), wo x und y Floats sind.
extern const cl_F_div_t floor2 (const cl_F& x, const cl_F& y); inline const cl_I floor1 (const cl_F& x, const cl_F& y) { return floor1(x/y); }
// ceiling2(x,y) liefert (ceiling x y), wo x und y Floats sind.
extern const cl_F_div_t ceiling2 (const cl_F& x, const cl_F& y); inline const cl_I ceiling1 (const cl_F& x, const cl_F& y) { return ceiling1(x/y); }
// truncate2(x,y) liefert (truncate x y), wo x und y Floats sind.
extern const cl_F_div_t truncate2 (const cl_F& x, const cl_F& y); inline const cl_I truncate1 (const cl_F& x, const cl_F& y) { return truncate1(x/y); }
// round2(x,y) liefert (round x y), wo x und y Floats sind.
extern const cl_F_div_t round2 (const cl_F& x, const cl_F& y); inline const cl_I round1 (const cl_F& x, const cl_F& y) { return round1(x/y); }
// Return type for decode_float:
struct decoded_float { cl_F mantissa; cl_I exponent; cl_F sign; // Constructor.
decoded_float () {} decoded_float (const cl_F& m, const cl_I& e, const cl_F& s) : mantissa(m), exponent(e), sign(s) {} };
// decode_float(x) liefert zu einem Float x: (decode-float x).
// x = 0.0 liefert (0.0, 0, 1.0).
// x = (-1)^s * 2^e * m liefert ((-1)^0 * 2^0 * m, e als Integer, (-1)^s).
extern const decoded_float decode_float (const cl_F& x);
// float_exponent(x) liefert zu einem Float x:
// den Exponenten von (decode-float x).
// x = 0.0 liefert 0.
// x = (-1)^s * 2^e * m liefert e.
extern sintE float_exponent (const cl_F& x);
// float_radix(x) liefert (float-radix x), wo x ein Float ist.
inline sintL float_radix (const cl_F& x) { (void)x; // unused x
return 2; }
// float_sign(x) liefert (float-sign x), wo x ein Float ist.
extern const cl_F float_sign (const cl_F& x);
// float_sign(x,y) liefert (float-sign x y), wo x und y Floats sind.
extern const cl_F float_sign (const cl_F& x, const cl_F& y);
// float_digits(x) liefert (float-digits x), wo x ein Float ist.
// < ergebnis: ein uintC >0
extern uintC float_digits (const cl_F& x);
// float_precision(x) liefert (float-precision x), wo x ein Float ist.
// < ergebnis: ein uintC >=0
extern uintC float_precision (const cl_F& x);
// Returns the floating point format of a float.
inline float_format_t float_format (const cl_F& x) { return (float_format_t) float_digits(x); }
// integer_decode_float(x) liefert zu einem Float x: (integer-decode-float x).
// x = 0.0 liefert (0, 0, 1).
// x = (-1)^s * 2^e * m bei Float-Precision p liefert
// (Mantisse 2^p * m als Integer, e-p als Integer, (-1)^s als Fixnum).
extern const cl_idecoded_float integer_decode_float (const cl_F& x);
// rational(x) liefert (rational x), wo x ein Float ist.
extern const cl_RA rational (const cl_F& x);
// scale_float(x,delta) liefert x*2^delta, wo x ein Float ist.
extern const cl_F scale_float (const cl_F& x, sintC delta); extern const cl_F scale_float (const cl_F& x, const cl_I& delta);
// max(x,y) liefert (max x y), wo x und y Floats sind.
extern const cl_F max (const cl_F& x, const cl_F& y);
// min(x,y) liefert (min x y), wo x und y Floats sind.
extern const cl_F min (const cl_F& x, const cl_F& y);
// signum(x) liefert (signum x), wo x ein Float ist.
extern const cl_F signum (const cl_F& x);
// Returns the largest (most positive) floating point number in float format f.
extern const cl_F most_positive_float (float_format_t f);
// Returns the smallest (most negative) floating point number in float format f.
extern const cl_F most_negative_float (float_format_t f);
// Returns the least positive floating point number (i.e. > 0 but closest to 0)
// in float format f.
extern const cl_F least_positive_float (float_format_t f);
// Returns the least negative floating point number (i.e. < 0 but closest to 0)
// in float format f.
extern const cl_F least_negative_float (float_format_t f);
// Returns the smallest floating point number e > 0 such that 1+e != 1.
extern const cl_F float_epsilon (float_format_t f);
// Returns the smallest floating point number e > 0 such that 1-e != 1.
extern const cl_F float_negative_epsilon (float_format_t f);
// Konversion zu einem C "float".
extern float float_approx (const cl_F& x);
// Konversion zu einem C "double".
extern double double_approx (const cl_F& x);
// Transcendental functions
// pi(y) liefert die Zahl pi im selben Float-Format wie y.
// > y: ein Float
extern const cl_F pi (const cl_F& y);
// pi(y) liefert die Zahl pi im Float-Format f.
// > f: eine Float-Format-Spezifikation
extern const cl_F pi (float_format_t f);
// pi() liefert die Zahl pi im Default-Float-Format.
extern const cl_F pi (void);
// sin(x) liefert den Sinus (sin x) eines Float x.
extern const cl_F sin (const cl_F& x);
// cos(x) liefert den Cosinus (cos x) eines Float x.
extern const cl_F cos (const cl_F& x);
// Return type for cos_sin():
struct cos_sin_t { cl_R cos; cl_R sin; // Constructor:
cos_sin_t () {} cos_sin_t (const cl_R& u, const cl_R& v) : cos (u), sin (v) {} };
// cos_sin(x) liefert ((cos x),(sin x)), beide Werte.
extern const cos_sin_t cos_sin (const cl_F& x);
// tan(x) liefert den Tangens (tan x) eines Float x.
extern const cl_F tan (const cl_F& x);
// exp1(y) liefert die Zahl e = exp(1) im selben Float-Format wie y.
// > y: ein Float
extern const cl_F exp1 (const cl_F& y);
// exp1(y) liefert die Zahl e = exp(1) im Float-Format f.
// > f: eine Float-Format-Spezifikation
extern const cl_F exp1 (float_format_t f);
// exp1() liefert die Zahl e = exp(1) im Default-Float-Format.
extern const cl_F exp1 (void);
// ln(x) liefert zu einem Float x>0 die Zahl ln(x).
extern const cl_F ln (const cl_F& x); // Spezialfall: x Long-Float -> Ergebnis Long-Float
inline const cl_LF ln (const cl_LF& x) { return The(cl_LF)(ln(The(cl_F)(x))); }
// exp(x) liefert zu einem Float x die Zahl exp(x).
extern const cl_F exp (const cl_F& x);
// sinh(x) liefert zu einem Float x die Zahl sinh(x).
extern const cl_F sinh (const cl_F& x);
// cosh(x) liefert zu einem Float x die Zahl cosh(x).
extern const cl_F cosh (const cl_F& x);
// Return type for cosh_sinh():
struct cosh_sinh_t { cl_R cosh; cl_R sinh; // Constructor:
cosh_sinh_t () {} cosh_sinh_t (const cl_R& u, const cl_R& v) : cosh (u), sinh (v) {} };
// cosh_sinh(x) liefert ((cosh x),(sinh x)), beide Werte.
extern const cosh_sinh_t cosh_sinh (const cl_F& x);
// tanh(x) liefert zu einem Float x die Zahl tanh(x).
extern const cl_F tanh (const cl_F& x);
// eulerconst(y) liefert die Eulersche Konstante
// im selben Float-Format wie y.
// > y: ein Float
extern const cl_F eulerconst (const cl_F& y);
// eulerconst(y) liefert die Eulersche Konstante im Float-Format f.
// > f: eine Float-Format-Spezifikation
extern const cl_F eulerconst (float_format_t f);
// eulerconst() liefert die Eulersche Konstante im Default-Float-Format.
extern const cl_F eulerconst (void);
// catalanconst(y) liefert die Catalansche Konstante
// im selben Float-Format wie y.
// > y: ein Float
extern const cl_F catalanconst (const cl_F& y);
// catalanconst(y) liefert die Catalansche Konstante im Float-Format f.
// > f: eine Float-Format-Spezifikation
extern const cl_F catalanconst (float_format_t f);
// catalanconst() liefert die Catalansche Konstante im Default-Float-Format.
extern const cl_F catalanconst (void);
// zeta(s) returns the Riemann zeta function at s>1.
extern const cl_F zeta (int s, const cl_F& y); extern const cl_F zeta (int s, float_format_t f); extern const cl_F zeta (int s);
// random_F(randomstate,n) liefert zu einem Float n>0 ein zufälliges
// Float x mit 0 <= x < n.
// > randomstate: ein Random-State, wird verändert
extern const cl_F random_F (random_state& randomstate, const cl_F& n);
inline const cl_F random_F (const cl_F& n) { return random_F(default_random_state,n); }
// This could be optimized to use in-place operations.
inline cl_F& operator+= (cl_F& x, const cl_F& y) { return x = x + y; } inline cl_F& operator+= (cl_F& x, const float y) { return x = x + y; } inline cl_F& operator+= (cl_F& x, const double y) { return x = x + y; } inline cl_F& operator++ /* prefix */ (cl_F& x) { return x = plus1(x); } inline void operator++ /* postfix */ (cl_F& x, int dummy) { (void)dummy; x = plus1(x); } inline cl_F& operator-= (cl_F& x, const cl_F& y) { return x = x - y; } inline cl_F& operator-= (cl_F& x, const float y) { return x = x - y; } inline cl_F& operator-= (cl_F& x, const double y) { return x = x - y; } inline cl_F& operator-- /* prefix */ (cl_F& x) { return x = minus1(x); } inline void operator-- /* postfix */ (cl_F& x, int dummy) { (void)dummy; x = minus1(x); } inline cl_F& operator*= (cl_F& x, const cl_F& y) { return x = x * y; } inline cl_F& operator*= (cl_F& x, const float y) { return x = x * y; } inline cl_F& operator*= (cl_F& x, const double y) { return x = x * y; } inline cl_F& operator/= (cl_F& x, const cl_F& y) { return x = x / y; } inline cl_F& operator/= (cl_F& x, const float y) { return x = x / y; } inline cl_F& operator/= (cl_F& x, const double y) { return x = x / y; }
// Thrown when a floating-point exception occurs.
class floating_point_exception : public runtime_exception { public: explicit floating_point_exception(const std::string & what) : runtime_exception(what) {} };
// Thrown when NaN occurs.
class floating_point_nan_exception : public floating_point_exception { public: floating_point_nan_exception(); };
// Thrown when overflow occurs.
class floating_point_overflow_exception : public floating_point_exception { public: floating_point_overflow_exception(); };
// Thrown when underflow occurs.
class floating_point_underflow_exception : public floating_point_exception { public: floating_point_underflow_exception(); };
// If this is true, floating point underflow returns zero instead of throwing an exception.
extern bool cl_inhibit_floating_point_underflow;
} // namespace cln
#endif /* _CL_FLOAT_H */
|